Subtle differences in the appearance of hippocampal layers differentiate preclinical AD from healthy aging: Neuroimaging / differential diagnosis. (7th December 2020)

• Record Type: Journal Article
• Authors: Xie, Long; Das, Sandhitsu R.; Wisse, Laura; Ittyerah, Ranjit; Vergnet, Nicolas; de Flores, Robin; Wolk, David A.; Yushkevich, Paul A.
• Is Part Of: Alzheimer's & dementia. Volume 16(2020)Supplement 4
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-12-07
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.045050
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Degeneration of the stratum radiatum lacunosum moleculare (SRLM), located between the cornu ammonis (CA)/subiculum and dentate gyrus (DG) subfields of the hippocampus (Figure 1) and containing apical dendrites of CA pyramidal cells, may be an early feature of AD and can be observed in postmortem studies (Adler et al, 2018). The pathological change in SRLM may produce subtle alteration of image signal on high‐resolution in vivo MRI. We hypothesize that convolutional neural networks (CNNs) can detect such alterations in preclinical AD, yielding a more sensitive biomarker than conventional morphological measurements. Methods: Cross‐sectional 0.4x0.4x2.0mm 3 T2‐weighted MRI of 285 cognitively normal controls from ADNI were included. The dataset was dichotomized according to amyloid‐ß (Aß) status (+ or ‐) based on PET and divided into training (48%), validation (12%) and test sets (40%, Table 1). Volumes of medial temporal lobe subregions (CA1‐3, DG, subiculum, entorhinal cortex, Brodmann areas 35/36, parahippocampal cortex) were extracted automatically using ASHS (Yushkevich et al., 2015). Small image patches (∼6x6x6 mm 3 ) centered on the SRLM were extracted. A CNN was trained to discriminate patches as Aß+ or Aß‐. The output of the last CNN layer was averaged over all SRLM patches to generate a CNN‐based score for each subject in the test set. ANCOVA with age, sex, education and intracranial volume as covariates was used to test for significance. Results: We observed a significant difference in CNN score between Ab+ and Ab‐ controls in the test set (Table 2). None of the conventional volume measures show statistical significance except a trend level effect in parahippcampal cortex opposite the expected result (greater atrophy in Aß+). Conclusions: In this proof‐of‐concept study, we demonstrated that our novel CNN‐based score is more sensitive to Aß status than standard medial temporal lobe volume measurements. Since the image patches only cover a small area around the SRLM, the score may reflect subtle change of the SRLM in preclinical AD. Better understanding the linkage of SRLM appearance and underlying pathology is required and will be investigated in future work using a dataset of antemortem and postmortem imaging in the same subjects.